The invention discloses a personalized multi-view federal recommendation system, which comprises a central server and a plurality of user clients, and any user client comprises a training module and a prediction module; wherein the training module comprises a data distribution sub-module, a gradient calculation sub-module, a gradient aggregation sub-module, a model updating sub-module, a model fine tuning sub-module, a user data warehouse and an article data warehouse which cooperate with one another to complete execution of a training algorithm, and a user sub-model and an article sub-model are obtained; and the prediction module comprises a semantic calculation sub-module, an interactive calculation sub-module, a probability aggregation sub-module, a probability sorting sub-module, a recommendation output sub-module, a user model warehouse and an article model warehouse which cooperate with one another to complete execution of a prediction algorithm and obtain a recommended article sequence corresponding to any user client. According to the method, the scene adaptability is higher, the feature mining of the underlying model is deeper, the data source covered by the original input is wider, and the localization fine tuning of the global model is better.